US 7634710 B2 Abstract Embodiments of a method and apparatus for decoding signals are disclosed. The method includes receiving modulated signals, generating bits representing the signals, and associated reliability of each bit. The method further includes executing a first stage of decoding the bits using a first component code, and simultaneously executing the first stage of decoding again using a second component code, and executing a second stage of decoding using the first component code. The first and second stages of decoding are used to generate the bit stream. Another method includes receiving modulated signals, generating bits representing the signals, and associated reliability of each bit. The method further includes executing a first stage of N stages for decoding the bits, the first stage using a first of M component codes, and simultaneously executing a plurality of the N stages of decoding, each of the plurality of N stages using a different one of the M component codes. The plurality of N stages of decoding are used to generate the bit stream.
Claims(16) 1. A method of decoding signals to generate a received bit stream, comprising:
receiving modulated signals;
generating bits representing the modulated signals, and beliefs representing an associated reliability of each bit;
a bit node computation block receiving the bits and associated beliefs, and generating a plurality of bit node messages;
a plurality of M serially-connected pipeline stages receiving the bit node messages and after M decoding cycles, generating a plurality of check node messages once per decoding cycle, comprising;
for each iteration cycle, each of the M serially-connected pipeline stages performing check node computations using all of J component codes, wherein each one of the M serially-connected pipeline stages performs check node computations once per decoding cycle using a single component code that is different than component codes used for all other of the M serially-connected pipeline stages, wherein J is at least as great as M, and wherein each iteration includes M decoding cycles.
2. The method of
every decoding cycle after M decoding cycles, the bit node computation block updating beliefs of the received bits using check node messages of a single one of the component codes per decoding cycle, wherein the single one component of the component codes changes for each decoding cycle according to a specific order of the component codes.
3. The method of
every decoding cycle, updating all bit node messages of the received bits using check node to bit node messages of a single one of the component codes per decoding cycle, wherein the single one component of the component codes changes for each decoding cycle according to a specific order of the component codes.
4. The method of
5. The method of
6. The method of
7. The method of
8. The method of
9. The method of
10. The method of
a first sub-block of the first stage re-ordering bit node to check node messages, enabling processing of the a second sub-block of the first stage to remain static as the first stage cycles through processing using the J component codes.
11. The method of
12. The method of
13. The method of
14. The method of
15. The method of
16. The method of
Description This application is a continuation-in-part (CIP) of patent application Ser. No. 11/090,109, filed Mar. 25, 2005 now U.S. Pat. No. 7,461,328, and herein incorporated by reference. The invention relates generally to network communications. More particularly, the invention relates to a method and apparatus for efficient decoding. High-speed networks are continually evolving. The evolution includes a continuing advancement in the operational speed of the networks. The network implementation of choice that has emerged is Ethernet networks physically connected over unshielded twisted pair wiring. Ethernet in its 10/100BASE-T form is one of the most prevalent high speed LANs (local area network) for providing connectivity between personal computers, workstations and servers. High-speed LAN technologies include 100BASE-T (Fast Ethernet) and 1000BASE-T (Gigabit Ethernet). Fast Ethernet technology has provided a smooth evolution from 10 Megabits per second (Mbps) performance of 10BASE-T to the 100 Mbps performance of 100BASE-T. Gigabit Ethernet provides 1 Gigabit per second (Gbps) bandwidth with essentially the simplicity of Ethernet. There is a desire to increase operating performance of Ethernet to even greater data rates. An implementation of high speed Ethernet networks includes simultaneous, full bandwidth transmission, in both directions (termed full duplex), within a selected frequency band. When configured to transmit in full duplex mode, Ethernet line cards are generally required to have transmitter and receiver sections of an Ethernet transceiver connected to each other in a parallel configuration to allow both the transmitter and receiver sections to be connected to the same twisted wiring pair for each of four pairs. The twisted pair LAN connections Information such as video, audio and data, is communicated over the networks as binary values. More specifically, the information is conveyed as zeros and ones. During transmission and processing of the information, errors can be unintentionally introduced. That is, a zero may be changed to a one, or vice versa. High speed networks, such as Gigabit Ethernet are more susceptible to these errors than slower speed networks because of the increased interference. To provide a mechanism to check for errors and in some cases to correct errors, binary data can be coded to introduce carefully designed redundancy. Coding of a unit of data produces what is commonly referred to as a codeword. Because of its redundancy, a codeword often includes more bits that the input unit of data from which the codeword was produced. When signals arising from transmitted codewords are received or processed, the redundant information included in the codeword as observed in the signal can be used to identify and/or correct errors in transmitted data. A class of codes referred to as low density parity check (LDPC), has emerged for use in data transmission. LDPC codes are coding schemes that use iterative decoders. It has been shown that LDPC codes can provide very good error correction for large codewords. Longer codewords are often more reliable at error detection and correction due to the coding interaction over a larger number of bits. The use of longer codewords can be beneficial by increasing the ability to detect and correct errors. However, a difficulty in adopting LDPC code that include long codewords is the complexity of implementing the code. LDPC coding using long codewords are significantly more complex than traditionally used coding systems such as convolution codes and Reed-Solomon codes. The added complexity requires additional power. It is desirable to have an apparatus and method for coding and decoding high speed transmission data that is complex, but yet dissipates low power. An embodiment of the invention includes a method of decoding signals to generate a received bit stream. The method includes receiving modulated signals, and generating bits representing the signals, and beliefs representing an associated reliability of each bit. A bit node computation block receives the bits and associated beliefs, and generates a plurality of bit node messages. A plurality of M serially-connected pipeline stages receives the bit node messages and after M decoding cycles, and generates a plurality of check node messages once per decoding cycle. For each iteration cycle, each of the M serially-connected pipeline stages perform check node computations using all of J component codes, wherein each one of the M serially-connected pipeline stages performs check node computations once per decoding cycle using a single component code that is different that component codes used for all other of the M serially-connected pipeline stages, wherein J is at least as great as M, and wherein each iteration includes M decoding cycles. Another embodiment includes a decoder. The decoder includes means for generating bits representing received signals, and beliefs representing an associated reliability of each bit. A bit node computation block receives the bits and associated beliefs, and generates a plurality of bit node messages. A plurality of M serially-connected pipeline stages receive the bit node messages and after M decoding cycles, and generate a plurality of check node messages once per decoding cycle, wherein for each iteration cycle, each of the M serially-connected pipeline stages performs check node computations using all of J component codes, wherein each one of the M serially-connected pipeline stages performs check node computations once per decoding cycle using a single component code that is different that component codes used for all other of the M serially-connected pipeline stages, wherein J is at least as great as M, and wherein each iteration includes M decoding cycles. Other aspects and advantages of the present invention will become apparent from the following detailed description, taken in conjunction with the accompanying drawings, illustrating by way of example the principles of the invention. As shown in the drawings for purposes of illustration, the invention is embodied in an apparatus and method for high performance decoding of a bit stream. The decoding can be implemented for low-power dissipation, and/or low latency. Decoding of a received bit stream can include multiplying blocks of the bit stream (code words) by a parity matrix. A resulting parity check vector can provide an estimation of the integrity of the received bit stream. An exemplary parity matrix can be given as:
An exemplary codeword that includes 7 bits of the data stream can be given as:
A parity check vector is determined by multiplying the codeword with the parity check matrix. More specifically, the parity check vector can be give as:
For this exemplary parity check vector, a first parity condition can be satisfied if x1+x2+x5=0, a second parity condition can be satisfied if x2+x3+x6=0, and a third parity condition can be satisfied if x3+x4+x7=0. The parity condition provides an indication of the integrity of the bits of the codeword of the received bit stream. The parity matrix and parity check vector provided here are for illustration purposes only. Tanner graphs (like the one shown in Decoding Phases The above-described SPA decoding includes two decoding phases per decoding iteration. The two phases include the bit node calculations and the check node calculations. The hardware dedicated to the bit node calculations is idle while the hardware that is dedicated to the check nodes is performing check node calculations, and the hardware that is dedicated to the check node calculations is idle while the hardware that is dedicated to the bit nodes is performing bit node calculations. Therefore, hardware executing the decoding is not fully utilized, and therefore, less efficient than desired. An iteration of decoding can be defined as including the bit node phase and the check node phase. The number of iterations for decoding received signals can be limited, or unlimited depending upon the decoding system. In practice, generally a cap is put on the maximum number of iterations. Early termination refers to a condition in which an iterative decoding algorithm is stopped before a maximum number of iterations have been executed. To achieve a desired bit error rate, a minimum number of iterations must be allocated to the decoding. Rate of Decoding SPA decoding is rate limited by the number of iterations, and the time required to execute each phase of decoding. More specifically, the rate of decoding can be estimated as 1/(2iT) where i is the number of iterations, and T is the time required for each decoding phase of the SPA decoding. The maximum number of iterations can be dependent upon several different constraints, but for the descriptive purposes, it is assumed to be fixed. With present decoding implementations, the rate is limited by the estimate (bounds) as just described. A Standard SPA Decoder A code can be viewed as a bipartite graph, such as the previously described Tanner Graph. The graph includes two types of nodes. Bit nodes correspond with each bit of a codeword. Check nodes correspond to each row of a parity check matrix. Associated with each bit node is a number that reflects the reliability by the decoder on a value of the bit. For a bit b, the number is referred to as Belief(b). For example, if Belief(b)=−∞ then b is certainly 1. If Belief(b)=∞, then b is certainly 0. If Belief(b)=0, then b is equally likely to be either a 0 and a 1. The SPA messages are passed between the check nodes and the bit nodes. Each message includes a real number. At any iteration of the decoding, these messages are used to update the belief values for the bit nodes. Each iteration of the decoding process can be divided into two phases. More specifically, the decoding process includes the bit node computations and the check node computations. Bit Node Computation Given the messages from each check node to bit node, each message from each bit node to connected check nodes is computed. Next, Belief values for each bit node are updated. That is:
In the first iteration these messages are all zero. Check Node Computation For each embodiment, the check node computation is based on a Gallagher function Φ. The Gallagher function is a well know function in decoding. More specifically,
Disadvantages of SPA A fundamental disadvantage of SPA is a requirement of knowledge of all Messages (b Advantages of Described Embodiments Over SPA The described embodiments provide advantages over SPA. More specifically, the bit nodes and the check nodes of the described embodiments are updated simultaneously, and do not require check node circuitry to be idle while waiting for all bit node or check node computations. At every stage of a pipeline decoder, only the set of check nodes in the component code processed by that stage are updated. A Pipeline Decoder Component Codes The decoding of Component Code Selection Each set of check nodes defines a corresponding component code. Various conditions and parameters can be used to select the component codes. As previously described, each of the component codes can be a subset of rows of the parity matrix. Component code selections can be made so that characteristics of the parity check matrix of the component codes are maintained. For example, the component codes can be selected so that column weights of a parity check matrix of the component codes are non-zero. This can be an advantageous selection because this selection avoids error patterns affecting bits corresponding to zero weight columns of the parity check matrix, from slowing down convergence of the decoding algorithm or process. Zero weight parity check matrix columns can cause bits being decoded to become stuck on incorrect values. The component codes can be selected so that each component code comprises a regular LDPC, and column weights of a parity matrix of the component codes are a pre-selected value. Making the column weights of all component codes of a common pre-selected value generally results in simpler hardware configurations required to implement the decoding. Note that a regular LDPC is an LDPC in which the column weights of the parity check matrix are all the same. Another embodiment includes column weights of a parity matrix of the component codes having pre-selected values that are not all the same. Generally, however, this embodiment leads to less efficient hardware implementations. When the pre-selected values are the same, the hardware implementing the decoding is typically more structured. That is, the hardware of each stage is more repetitive, and therefore, easier to implement. The component codes can be selected so that a parity check matrix of the component codes has column weights of one. This configuration generally provides the maximum number of component codes and the maximum number of stages. This configuration is most ideal for maximum clock rates, and therefore, greater throughput than other decoding configurations. Mathematical Representations Decoding of the codes is highly dependent on the block permutation structure of the parity matrix of the codes. The following is a mathematical formulation of the decoding, having notation defined as follows. The parity check matrix of
Defining B The vector (X The vector V The mapping Φ: R→R, that defines the so-called Gallager function in the standard SPA. This mapping is generalized in a component-by-component manner to vectors in higher dimensions. The main structural property that is invoked here includes observing that the message (X From the above observation it follows that the bit node messages can be obtained by:
_{i})^{c}=Φ^{−1}(Σ_{j≠i}Φ(V _{i} ^{c}))Finally the updated rules for beliefs are: B _{i} =B _{i} −P ^{−1} _{c,i}(( X _{i})^{c}−(X _{i})^{c})Pipeline Decoding The decoding embodiments of Latency The pipeline decoding structure of Other Pipeline Decoding Embodiments A bit node computation block The bit node computation block The bit node computation block The decoder shown in An embodiment additionally includes every decoding cycle after M decoding cycles, the bit node computation block updating beliefs of the received bits using check node messages of a single one of the component codes per decoding cycle, wherein the single one component of the component codes changes for each decoding cycle according to a specific order of the component codes. The specific order can include a randomly selected order of the set of component codes. Additionally, an embodiment includes every decoding cycle, updating all bit node messages of the received bits using check node to bit node messages of a single one of the component codes per decoding cycle, wherein the single one component of the component codes changes for each decoding cycle according to a specific order of the component codes. For an embodiment the bit node computation block generates a decoded received bit stream once for every N iterations, wherein N is a pre-selected number of iterations. For a specific embodiment, N is dynamically selected depending on whether all parity checks of the codes are satisfied. More specifically, once all parity checks have been matched, the decoding algorithm is terminated. For an embodiment, the check node computations of each of the M serially-connected pipeline stages, includes different functionality than all other of the M serially connected pipeline stages. For a more specific embodiment, for each cycle of decoding, each of M−1 of the M serially-connected pipeline stage executes one step of check node computations and provides intermediate results to a next stage of the M serially-connected pipeline stages. For an embodiment, for each iteration of decoding, each stage of the M serially-connected pipeline stages of decoding using updated bit node messages including all previous iterations. For an embodiment, the column weights of a parity check matrix of the component codes are non-zero. An embodiment includes the parity check matrix including elements of ones and zeroes. A non-zero column weight means at least one of the elements of a column of the matrix is one. The column weight can be defined as the number of non-zero elements of the column. An embodiment includes configuring a first of the M serially-connected pipeline stages with a specific component code, and configuring a last of the M serially-connected pipeline stages with another specific component code. A implementation of this embodiment includes a first sub-block of the first stage re-ordering bit node to check node messages, enabling processing of the a second sub-block of the first stage to remain static as the first stage cycles through processing using the J component codes. For an embodiment, the re-ordering of the bit node to check node messages changes every decoding cycle. Additionally, an embodiment includes a second sub-block of the last (Mth) stage re-ordering check node to bit node messages, enabling processing of the a first sub-block of the last stage to remain static as the last stage cycles through processing using the J component codes. For an embodiment, the re-ordering of the check node to bit node messages changes every decoding cycle. An embodiment includes a decoder. The decoder includes means for generating bits representing received signals, and beliefs representing a channel reliability of each bit. A bit node computation block receives the bits and associated beliefs, and generates a plurality of bit node messages and beliefs for each bit. Check node messages are initialized to zero at the beginning of the decoding. A plurality of M serially-connected pipeline stages receive the bit node messages and after M decoding cycles, generating a plurality of check node messages once per decoding cycle, wherein for each iteration cycle, each of the M serially-connected pipeline stages performs check node computations using all of J component codes, wherein each one of the M serially-connected pipeline stages performs check node computations once per decoding cycle using a single component code that is different that component codes used for all other of the M serially-connected pipeline stages, wherein J is at least as great as M, and wherein each iteration includes M decoding cycles. A Network of Devices Although specific embodiments of the invention have been described and illustrated, the invention is not to be limited to the specific forms or arrangements of parts so described and illustrated. The invention is limited only by the appended claims. Patent Citations
Non-Patent Citations
Classifications
Legal Events
Rotate |